Alternative Fecal Indicators and Their Empirical Relationships with Enteric Viruses, Salmonella enterica, and Pseudomonas aeruginosa in Surface Waters of a Tropical Urban Catchment

Abstract

The suitability of traditional microbial indicators (i.e., Escherichia coli and enterococci) has been challenged due to the lack of correlation with pathogens and evidence of possible regrowth in the natural environment. In this study, the relationships between alternative microbial indicators of potential human fecal contamination (Bacteroides thetaiotaomicron, Methanobrevibacter smithii, human polyomaviruses [HPyVs], and F+ and somatic coliphages) and pathogens (Salmonella spp., Pseudomonas aeruginosa, rotavirus, astrovirus, norovirus GI, norovirus GII, and adenovirus) were compared with those of traditional microbial indicators, as well as environmental parameters (temperature, conductivity, salinity, pH, dissolved oxygen, total organic carbon, total suspended solids, turbidity, total nitrogen, and total phosphorus). Water samples were collected from surface waters of urban catchments in Singapore. Salmonella and P. aeruginosa had significant positive correlations with most of the microbial indicators, especially E. coli and enterococci. Norovirus GII showed moderately strong positive correlations with most of the microbial indicators, except for HPyVs and coliphages. In general, high geometric means and significant correlations between human-specific markers and pathogens suggest the possibility of sewage contamination in some areas. The simultaneous detection of human-specific markers (i.e., B. thetaiotaomicron, M. smithii, and HPyVs) with E. coli and enterococcus supports the likelihood of recent fecal contamination, since the human-specific markers are unable to regrow in natural surface waters. Multiple-linear-regression results further confirm that the inclusion of M. smithii and HPyVs, together with traditional indicators, would better predict the occurrence of pathogens. Further study is needed to determine the applicability of such models to different geographical locations and environmental conditions.

INTRODUCTION

Indicator bacteria, such as Escherichia coli and enterococci (ENT), have been used extensively in past decades to indicate the presence of human enteric pathogens because of the great diversity of pathogens, the large volumes required, and the difficulty in concentrating the typically low concentrations of pathogens from water samples. In Singapore, microbial water quality guidelines for recreational freshwater currently target enterococci (less than 200 counts per 100 ml for 95% of the time).

Despite the wide use of fecal indicator bacteria in microbial water quality monitoring, their suitability to represent a level of health risk is still being questioned. One of the reasons is the lack of data correlating the presence and concentrations of the indicators with the presence of disease-associated pathogens. With improvements in microbiological test methods and bacterial species identification, there is growing evidence that fecal indicator bacteria, including E. coli and ENT, do not necessarily correlate well with some of the enteric pathogens, particularly viruses (1). There is also evidence that E. coli and ENT are able to survive, grow, and establish populations in natural environments, such as freshwater lakes and streams, soils, and sediments (2). Thus, there is a need to examine other possible improved indicators or markers for assessing human fecal contamination.

In recent years, human-associated markers have been suggested as alternatives to indicate the presence of human enteric pathogens. Many waterborne pathogens, particularly viruses, are found in human fecal samples (3). As such, human sewage contamination may pose higher risks to human health than animal wastes. Bacteroides thetaiotaomicron, human polyomaviruses (HPyVs), and Methanobrevibacter smithii are some of the potential markers that have shown high specificity for human contamination (4–6). Many studies have examined the correlation between these alternative markers and human pathogens in natural aquatic environments. However, there is no universal trend. For example, in a study by McQuaig et al., the presence of adenoviruses was correlated with HPyVs and human-associated Bacteroides sp. strain HF183 at Doheny Beach on the west coast of the United States (7). Other studies have shown that human-associated Bacteroides and HPyVs demonstrated persistence similar to that of pathogens such as adenovirus and enterovirus (8).

A comprehensive review of correlations between markers and pathogens in surface waters has been provided by Harwood et al. (9). The inconsistent results of correlations between human-specific or fecal markers and human pathogens reported by different studies could be due to different environmental conditions. Temperature is one factor that plays an important role in the survival of microorganisms. For instance, bacterial indicators were found to survive longer and even to multiply in tropical environmental waters (2). In addition to temperature, other environmental factors, such as turbidity and nutrients, have been shown to significantly impact correlation studies (10). The studies cited above focused on only a limited number of indicators and pathogens, making cross comparisons between various studies difficult. Here, we examined the occurrence of traditional fecal indicators (E. coli and ENT) and a number of alternatives and, in some cases, human-specific fecal indicators (F+ coliphage, somatic coliphage, HPyVs, M. smithii, and B. thetaiotaomicron), human pathogens (rotavirus, astrovirus, norovirus GI, norovirus GII, and adenovirus), and both frank and opportunistic bacterial pathogens, specifically, Salmonella spp. and Pseudomonas aeruginosa, in water samples collected from urban and nonurban catchments in tropical Singapore. Information is provided on a variety of microbial indicators and pathogens of concern simultaneously to facilitate comparison of the performances of traditional, as well as alternative and human-specific, fecal indicators. In addition to correlations between microorganisms, this study also evaluated the impacts of various physical and chemical environmental factors on the occurrence of indicators and pathogens. The data collected were used to derive regression models to predict the occurrence of pathogens in the tropical surface waters of Singapore. Ultimately, the detection of effective microbial indicators could reflect the source of human fecal pollution, which could contribute to quantitative microbial risk assessment and remediation actions.

MATERIALS AND METHODS

Site description.

In total, 148 water samples were collected from a highly urbanized catchment in the southern part of Singapore. The water from upstream catchments flows through commercial, residential, and light industrial areas and feeds into a reservoir via five main tributaries. Currently, this reservoir is the largest in Singapore, serving a catchment area of 10,000 ha. Apart from storing storm water for potable-water use after treatment, the reservoir provides flood control and is also used for recreational activities, including kayaking, water skiing, and boating.

An additional 20 samples were collected from drains and canals in different parts of Singapore to capture microbial water quality characteristics of subcatchments with different land uses, including agricultural, commercial, industrial, rural, and undeveloped land.

Due to the limited volume of some water samples, not all the target microorganisms were analyzed in all the samples. The exact numbers of samples analyzed for each microorganism are provided (see Table 2).

TABLE 2.

Frequencies of occurrence for pathogens and microbial indicators from all sites

Microorganism and test (units for geometric means/SD/detection limits)	No. of samples	% positive	Geometric meana	SD	Detection limit
E. coli, Colilert (MPN/100 ml)	148	99	803	13,580	0
E. coli (GC/100 ml)	103	100	1,309	70,983	60
ENT, Enterolert (MPN/100 ml)	148	98	118	5,212	0
ENT (GC/100 ml)	80	96	1,099	54,840	60
B. thetaiotaomicron (GC/100 ml)	109	35	69	2,886	60
M. smithii (GC/100 ml)	123	20	41	100	60
HPyVs (GC/100 ml)	138	46	70	2,010	24
F+ coliphage (PFU/100 ml)	83	94	27	228	0
Somatic coliphage (PFU/100 ml)	83	97	52	338	0
Salmonella spp. (CFU/100 ml)	95	86	13	28	0
P. aeruginosa (CFU/100 ml)	61	93	161	1,230	0
Rotavirus (GC/liter)	93	39	11	250	9
Astrovirus (GC/liter)	93	24	57	237	86
Norovirus GI (GC/liter)	93	20	7	99	9
Norovirus GII (GC/liter)	93	48	104	2,527	86
Adenovirus (GC/liter)	93	33	13	46	13

^aGeometric means were calculated by converting nondetections to half the lower detection limit.

Sample collection and measurement of physical-chemical parameters.

For the reservoir and its catchment, monthly sampling was conducted at each of the four tributaries in the upstream section of the catchment (sites 2 to 5, designated nonrecreational) and one location within the reservoir (site 1, designated recreational) from December 2011 to March 2012 and from July 2012 to April 2013 (Fig. 1). Water samples from each site were collected in 5-liter (for microbial indicators and bacterial pathogens) and 10-liter (for enteric viruses) sterile containers. For microbial indicators and bacterial pathogens, the volumes of water samples analyzed were 100 ml for each of the culture-based methods, 200 ml for quantitative PCR (qPCR) detection of bacterial indicators (4, 6, 11–13), and 500 ml (after adjusting the pH to 3.5 to promote better electrostatic interactions between HPyVs and filters) for human polyomavirus (5). From May to June 2012, 20 additional water samples were conducted at 12 other stations (sites 6 to 17) throughout Singapore (Fig. 1). All the samples collected were kept in a cold room and analyzed within 6 h of collection.

FIG 1.

Locations of sampling sites.

Physical-chemical parameters, including temperature, conductivity, salinity, pH, dissolved oxygen (DO) (HI9828 Multiparameter Meter; Hanna Instruments), total organic carbon (TOC) (TOC analyzer; Shimadzu), total suspended solids (TSS), turbidity, total nitrogen (TN), and total phosphorus (TP), were measured according to the American Public Health Association (APHA) standard methods (14).

Sample concentration and analysis.

Water samples were divided for analysis of seven different microbial indicators and several human pathogens. The conventional fecal indicators, E. coli and ENT, were quantified by analyzing 100-ml samples with Colilert and Enterolert test kits (IDEXX Laboratories, Inc., Westbrook, ME), respectively. E. coli and ENT, as well as human-specific microbial indicators, including B. thetaiotaomicron, M. smithii, and HPyVs, were evaluated using real-time (RT) qPCR, while F+ and somatic coliphages were enumerated using the single-agar-layer method (15). Salmonella spp. and P. aeruginosa were measured following APHA standard methods (14), while enteric viruses (i.e., rotavirus, astrovirus, norovirus GI, norovirus GII, and adenovirus) were analyzed by qPCR after concentrating water samples from 10 liters to 0.5 ml by first prefiltering them through 20-μm-pore-size filter paper, followed by tangential-flow ultrafiltration (Sartorus, Germany) with a 30-kDa membrane cassette and further centrifugal ultrafiltration with an Ultra-15 centrifugal tube with cutoff levels of 30 kDa (Amicon Merk, Germany) (16–20). After concentration, 120 μl of DNA was extracted using the PowerSoil DNA isolation kit (Mo Bio Laboratories, Inc., Carlsbad, CA) for analysis of indicators and 60 μl was extracted using a QIAamp Viral RNA minikit (Qiagen, Inc., Valencia, CA) for virus analysis, with storage at −20°C for later analysis.

All the primers and probes for qPCR are listed in Table 1. All the qPCRs were performed in 20 μl of reaction mixture containing 5 μl of extracted nucleic acid; 10 μl of FastStart Universal Probe Master (ROX) (Roche, Germany); and 5 μl of primers, probes, and nuclease-free reagent grade water (double-distilled water [ddH₂O]), according to previous studies (4–6, 11–13, 16–18, 20). The qPCR program for each microorganism also followed these studies (4–6, 11–13, 16–18, 20). Duplicate qPCR amplifications were performed in a StepOnePlus real-time PCR system. For each run, both positive (recombinant plasmid DNA [Integrated DNA Technologies, Inc.]) and negative (ddH₂O) controls were included. To minimize inhibition in qPCRs, inhibition tests were carried out using the TaqMan exogenous internal positive-control (IPC) kit. A sample was considered to be inhibited when the difference (IPC ΔC_T) between the internal positive-control cycle threshold (IPC C_T) values with and without sample DNA was above 3 C_T units (12). In this case, samples were diluted until the IPC ΔC_T value fell below 3 C_T units. In this study, the average IPC ΔC_T values were below 1.89 C_T units.

TABLE 1.

Oligonucleotide primers and probes used for qPCR measurements of microbial indicators and pathogens

Target microorganism	Sequence (5′–3′)a	Reference
E. coli (uidA gene)
Forward primer	GTG TGA TAT CTA CCC GCT TCG C	11
Reverse primer	AGA ACG GTT TGT GGT TAA TCA GGA
Probe	FAM-TCG GCA TCC GGT CAG TGG CAG T-TAMRA
ENT (23S rRNA gene)
Forward primer	GAG AAA T TC CAA ACG AAC T TG	11, 12
Reverse primer	CAG TGC TCT ACC TCC ATC ATT
Probe	FAM-TGG TTC TCT CCG AAA TAG CTT TAG GGC TA-TAMRA
B. thetaiotaomicron (alpha-mannanase gene)
Forward primer	CAT CGT TCG TCA GCA GTA ACA
Reverse primer	CCA AGA AAA AGG GAC AGT GG	4
Probe	FAM-ACC TGC TG-NFQ
M. smithii (nifH gene)
Forward primer	GAA AGC GGA GGT CCT GAA	6
Reverse primer	ACT GAA AAA CCT CCG CAA AC
Probe	FAM-CCG GAC GTG GTG TAA CAG TAG CTA-BHQ-1
HPyVs (T antigen)
Forward primer	ACT CTT TAG GTT CTT CTA CCT TT	5
Reverse primer	GGT GCC AAC CTA TGG AAC AG
Probe	FAM–TCA TCA CTG GCA ACC AT–MGBNFQ
Rotavirus (NSP3 gene)
Forward primer	ACC ATC TWC ACR TRA CCC TCT ATG AG	20
Reverse primer	GGT CAC ATA ACG CCC CTA TAG C
Probe	FAM-AGT TAA AAG CTA ACA CTG TCA AA-MGB
Astrovirus
Forward primer	GCT TCT GAT TAA ATC AAT TTT AA	16
Reverse primer	CCG AGT AGG ATC GAG GGT
Probe	FAM-CTT TTC TGT CTC TGT TTA GAT TAT TTT AAT CAC C-TAMRA
Adenovirus (hexon gene)
Forward primer	GGA CGC CTC GGA GTA CCT GAG	18
Reverse primer	ACG TGG GGT TTC TGA ACT TGT T
Probe	FAM-CTG GTG CAG TTC GCC CGT GCC A-BHQ
Norovirus GI (ORF1-ORF2 junction regions)
Forward primer	CGY TGG ATG CGN TTY CAT GA	17
Reverse primer	CTT AGA CGC CAT CAT CAT TYA C
Probe	FAM-AGA TYG CGA TCY CCT GTC CA-TAMRA
	FAM-AGA TCG CGG TCT CCT GTC CA-TAMRA
Norovirus GII (ORF1-ORF2 junction regions)
Forward primer	CAR GAR BCN ATG TTY AGR TGG ATG AG	17
Reverse primer	TCG ACG CCA TCT TCA TTC ACA
Probe	FAM-TGG GAG GGC GAT CGC AAT CT-TAMRA

^aMixed bases in the primers and probes are as follows: B, not A; N, any; R, A or G; W, A or T; Y, C or T. FAM, 6-carboxyfluorescein; TAMRA, 6-carboxytetramethylrhodamine.

Statistical analysis.

In this study, the U.S. Environmental Protection Agency (EPA)-recommended software, ProUCL version 5.0.00, was used to estimate data for nondetections by using the regression on order statistics method. The method was used to generate imputed values for nondetections. All statistical analysis was performed using SPSS version 18 (SPSS Inc., Chicago, IL) and Microsoft Excel 2010 (Microsoft Corporation, Redmond, WA). One-way analysis of variance (ANOVA) was used to compare means of microbial concentrations for the five sites at the reservoir catchment (i.e., sites 1 to 5). Spearman's rho correlation coefficients were calculated to determine the different relationships between pathogens and microbial-indicator concentrations, as well as the concentrations of environmental parameters. Different levels of correlation were defined (i.e., strong, r > 0.5; moderately strong, 0.3 < r < 0.5; and weak, r < 0.3). In addition, to further examine the relationships between the concentrations of norovirus GII and indicators, stepwise multiple linear regressions were performed. Model performance was assessed with an adjusted r² value, which represents the percentage of the variation that the model's independent variables describe (21).

RESULTS

Environmental parameters.

A total of 148 water samples were collected from surface waters in Singapore. The water temperature remained relatively constant, ranging from 27.2 to 32.9°C, while the pH ranged from 7.04 to 9.99. Other water quality parameters varied widely, including conductivity (154 to 1,690 μS; mean, 416 μS), salinity (96.2 to 864 ppm; mean, 244 ppm), DO (0.65 to 11.93 ppm; mean, 5.57 ppm), TSS (4 to 276 mg/liter; mean, 26 mg/liter), turbidity (3.7 to 664 nephelometric turbidity units [NTU]; mean, 41.2 NTU), TN (0.09 to 69.4 ppm; mean, 4.02 ppm), TP (0.011 to 1.9 ppm; mean, 0.291 ppm), and TOC (2.06 to 17.91 ppm; mean, 5.14 ppm). The antecedent dry period (i.e., the number of dry days before sampling time) was obtained from the National Environment Agency (NEA) of Singapore and ranged from 1 to 15 days with a mean value of 5 days.

Occurrence of microbial indicators and pathogens in recreational waters.

Among the four microbial indicators tested using culture-based methods, E. coli, ENT, and F+ and somatic coliphages were detected in more than 90% of samples, with geometric mean counts of 803 most probable number (MPN)/100 ml, 118 MPN/100 ml, 27 PFU/100 ml, and 52 PFU/100 ml, respectively (Table 2). qPCR results for E. coli and ENT were generally higher than those from the culture-based methods, with geometric mean counts of 1,309 gene copies (GC)/100 ml and 1,099 GC/100 ml, respectively. For the other human-specific microbial indicators, the positive samples detected by qPCR were all less than 50%. The geometric mean counts for B. thetaiotaomicron, M. smithii, and HPyVs were 69 GC/100 ml, 41 GC/100 ml, and 70 GC/100 ml, respectively.

Pathogens (including opportunistic ones) were detected in surface water samples using both culture and molecular detection methods. Salmonella spp. and P. aeruginosa were detected in 86% (n = 95) and 93% (n = 61) of the samples, with geometric mean counts of 13 CFU/100 ml and 161 CFU/100 ml, respectively. The pathogenic viruses were detected in less than 50% of the samples, with concentrations up to 1,800 GC/liter. Among the enteric viruses, norovirus GII was the most prevalent and was detected in 48% of the samples.

To compare different microorganisms from the five catchment sites (i.e., sites 1 to 5), the geometric means of each microorganism from the five sites were calculated (Table 3). Only samples from site 2 exceeded the Singapore ENT guideline of 200/100 ml for both culture-based and qPCR results (976 MPN/100 ml and 29,871 GC/100 ml, respectively), assuming 4 GC are equal to 1 cell equivalent (http://rrndb.umms.med.umich.edu). None of the human-specific indicators (i.e., B. thetaiotaomicron, M. smithii, and HPyVs) were detected at site 1. However, all four of the other catchment sites had human-specific markers, with concentrations ranging from 17 to 1,726 GC/100 ml, and elevated E. coli and ENT concentrations compared with site 1.

TABLE 3.

Detection of microorganisms at each sampling site

Microorganism and test	Geometric mean (% positive samples) at sitea:
	1	2	3	4	5
E. coli, Colilert (MPN/100 ml)	30 (100)	6,279 (96)	1,020(100)	454 (100)	1,655 (100)
E. coli (GC/100 ml)	151 (100)	26,772(100)	1,170 (100)	272 (100)	2,872 (100)
ENT, Enterolert (MPN/100 ml)	7 (92)	976 (100)	114 (100)	50 (100)	103 (100)
ENT (GC/100 ml)	109 (100)	29,871 (100)	730 (100)	207 (80)	744 (100)
B. thetaiotaomicron (GC/100 ml)	BDL (0)	1,726 (100)	29 (32)	33 (14)	38 (32)
M. smithii (GC/100 ml)	BDL (0)	90 (62)	36 (14)	BDL (0)	38 (24)
HPyVs (GC/100 ml)	BDL (0)	727 (96)	37 (58)	17 (33)	28 (42)
F+ coliphage (PFU/100 ml)	6 (85)	148 (100)	16 (100)	15 (100)	19 (92)
Somatic coliphage (PFU/100 ml)	10 (85)	189 (100)	99 (100)	46 (100)	57 (100)
Salmonella spp. (CFU/100 ml)	6 (68)	23 (89)	11 (89)	10 (95)	22 (89)
P. aeruginosa (CFU/100 ml)	29 (92)	393 (100)	256 (100)	106 (83)	349 (92)
Rotavirus (GC/liter)	10 (33)	19 (47)	11 (40)	9 (33)	11 (40)
Astrovirus (GC/liter)	65 (20)	28 (47)	54 (33)	45 (7)	49 (13)
Norovirus GI (GC/liter)	9 (27)	12 (40)	7 (20)	5 (7)	5 (7)
Norovirus GII (GC/liter)	156 (60)	253 (80)	69 (47)	55 (13)	79 (40)
Adenovirus (GC/liter)	12 (33)	24 (60)	12 (33)	11 (20)	9 (20)

^aBDL, below detection limit.

Correlations between pathogens, microbial indicators, and environmental parameters.

As shown in Table 4, there were significant positive correlations between all of the microbial indicators (the r value ranged from 0.379 to 0.84). The highest correlation (r = 0.84) was between E. coli and ENT from culture-based measurements. Among the environmental parameters, turbidity had strong positive correlations with all of the microbial indicators, where E. coli and HPyVs gave higher r values (>0.6). TN also showed strong positive correlations with almost all of the microbial indicators (r, 0.428 to 0.781), except for somatic coliphage.

TABLE 4.

Spearman rank correlations between different indicators

Indicator or parameter	Correlation (r)a
	E. coli, Colilert (MPN/100 ml)	E. coli (GC/100 ml)	ENT, Enterolert (MPN/100 ml)	ENT (GC/100 ml)	B. thetaiotaomicron (GC/100 ml)	M. smithii (GC/100 ml)	HPyVs (GC/100 ml)	F+ coliphage (PFU/100 ml)	S. coliphage (PFU/100 ml)
E. coli, Colilert (MPN/100 ml)
E. coli (GC/100 ml)	0.632b
ENT, Enterolert (MPN/100 ml)	0.840b	0.674b
ENT (GC/100 ml)	0.673b	0.836b	0.672c
B. thetaiotaomicron (GC/100 ml)	0.566b	0.775b	0.571b	0.802b
M. smithii (GC/100 ml)	0.549b	0.541b	0.482b	0.537b	0.615b
HPyVs (GC/100 ml)	0.608b	0.737b	0.663b	0.753b	0.722b	0.642b
F+ coliphage (PFU/100 ml)	0.588b	0.517c	0.571b	0.506c	0.652b	0.479b	0.531b
Somatic coliphage (PFU/100 ml)	0.581b	0.530c	0.497b	0.495c	0.560b	0.379b	0.394c	0.782b
Conductivity (μS/cm)
Salinity (ppm)
pH	−0.372b	−0.328c	−0.362b		−0.343b	−0.240c	−0.234c
DO (mg/liter)	−0.247c	−0.587b		−0.573b	−0.674b	−0.296c	−0.230c
TSS (mg/liter)	0.381b	0.407c					0.382b
Turbidity (NTU)	0.649b	0.524b	0.577b	0.397b	0.581b	0.408b	0.645b	0.299c	0.259c
TN (ppm)	0.587b	0.735b	0.723b	0.736b	0.725b	0.524b	0.781b	0.428b
TP (ppm)	0.360b		0.471b				0.428b	0.311c
TOC (ppm)	−0.263c		−0.308b
Temp (°C)	−0.292c	−0.477b
Antecedent dry period (days)	−0.478b		−0.450b

^aOnly significant correlations (P < 0.05 and P < 0.01) are shown. MPN, most probable number; GC, gene copies; PFU, plaque-forming units; ppm, part per million; NTU, nephelometric turbidity units.

^bCorrelation is significant (P < 0.01; 2-tailed t test).

^cCorrelation is significant (P < 0.05; 2-tailed t test).

Some of the pathogens had significant correlations (P < 0.05) with some of the microbial indicators and environmental parameters, except for rotaviruses and astrovirus, which did not correlate well with any parameter or indicator (Table 5).

TABLE 5.

Spearman rank correlations between pathogens, microbial indicators, and environmental parameters

Microbial indicator or parameter	Correlation (r)a
	Salmonella sp. (CFU/100 ml)	P. aeruginosa (CFU/100 ml)	Rotavirus (GC/liter)	Astrovirus (GC/liter)	Norovirus GI (GC/liter)	Norovirus GII (GC/liter)	Adenovirus (GC/liter)
E. coli, Colilert (MPN/100 ml)	0.643b	0.712b
E. coli (GC/100 ml)	0.566b	0.493b				0.453b	0.372c
ENT, Enterolert (MPN/100 ml)	0.627b	0.634b
ENT (GC/100 ml)	0.342c	0.499b				0.487b	0.637b
B. thetaiotaomicron (GC/100 ml)	0.376c	0.389b				0.421b
M. smithii (GC/100 ml)		0.400b			0.397b	0.330c
HPyVs (GC/100 ml)	0.351c	0.483b			0.440b		0.294c
F+ coliphage (PFU/100 ml)	0.448b
Somatic coliphage (PFU/100 ml)	0.439b	0.395b
Environmental parameters
Conductivity (μS)
Salinity (ppm)				−0.335c
pH	−0.380c
DO (mg/liter)	−0.486b	−0.754b			−0.360b		−0.347c
TSS (mg/liter)
Turbidity (NTU)	0.431b	0.642b
TN (ppm)	0.493b	0.728b			0.368b
TP (ppm)	0.384c	0.563c
TOC (ppm)				−0.360b
Temp (°C)
Antecedent dry period (days)

^aOnly significant correlations (P < 0.05 and P < 0.01) are shown.

^bCorrelation is significant (P < 0.01; 2-tailed t test).

^cCorrelation is significant (P < 0.05; 2-tailed t test).

Bacterial pathogens (i.e., Salmonella spp. and P. aeruginosa) had significant positive correlations with most of the microbial indicators (P < 0.05). Among them, E. coli and ENT had higher correlations with Salmonella spp. (the r value ranged from 0.627 to 0.643 for culture-based measurements of E. coli and ENT and was 0.566 for E. coli using qPCR), while E. coli, ENT, and HPyVs had higher correlations with P. aeruginosa (the r value ranged from 0.483 to 0.712). Among the viral pathogens, norovirus GII showed significant positive correlations (P < 0.05) with most of the microbial indicators (except HPyVs, F+ coliphage, and somatic coliphage). E. coli (qPCR), ENT (qPCR), and B. thetaiotaomicron had higher correlations with norovirus GII (r = 0.453, 0.487, and 0.421, respectively), while HPyVs correlated better with norovirus GI (r = 0.44). Adenovirus showed significantly better correlations with ENT (qPCR) (r = 0.637). However, neither rotavirus nor astrovirus correlated well with any of the microbial indicators (P > 0.05).

Between pathogens and environmental parameters, DO had significant negative correlations and TN had significant positive correlations with some of the pathogens; P. aeruginosa had the strongest correlation, with r values of −0.754 and 0.728, respectively. Turbidity and TP had positive correlations with Salmonella spp. and P. aeruginosa, while pH had a negative correlation (r = −0.38) with Salmonella spp. and TOC had a negative correlation (r = −0.36) with astrovirus.

Multiple-linear-regression analysis.

Since noroviruses are the single largest cause of documented outbreaks of recreational waterborne diseases among viruses (22) and Salmonella spp. and norovirus GII were among the most prevalent pathogens detected in our samples, their relationships with traditional microbial indicators and novel human-specific microbial indicators were compared. Multiple linear regressions were applied first to determine the relationships between these pathogens and E. coli and ENT. Subsequently, the relationships between selected pathogens and other indicators were studied by using stepwise multiple-linear-regression analysis. The models were created using data from all the sampling locations so that the models were not specific. All the models and their respective individual variables with significant P values were selected automatically by the software. As a result, different models were created to predict the concentrations of Salmonella spp. and norovirus GII using different indicators (Table 6).

TABLE 6.

Multiple linear regression models of Salmonella spp. and norovirus GII concentrations

Model	r2	Model significance	Indicator	Coefficient	95% confidence interval	Significancea	Constant
Salmonella spp.
1	0.426	0.001	ENT (Enterolert)	0.015	0.007 to 0.023	0.001	6.399
2	0.966	0.000	ENT (Enterolert)	0.048	0.044 to 0.053	0.000	−1.891
			HPyVs	−0.013	−0.017 to −0.009	0.000
			TN	8.027	2.829 to 13.226	0.004
Norovirus GII
1	0.153	0.02	E. coli (qPCR)	0.001	0 to 0.003	0.02	115.212
2	0.442	0.000	ENT (qPCR)	0.018	0.011 to 0.025	0.000	46.601
3	0.762	0.000	M. smithii	2.889	2.214 to 3.563	0.000	28.643
			HPyVs	−0.161	−0.299 to −0.023	0.001

^aOnly significances where the P value is <0.05 are shown.

The first few models in each pathogen group included only E. coli and ENT as variables, with r² values ranging from 0.153 to 0.442. This suggests that the concentration of each pathogen had only 15.3% to 44.2% of the data variation explained by E. coli and ENT. When additional human-specific microbial indicators (such as M. smithii and HPyVs) and TN were included in the analysis, the resulting model (i.e., the last model in each pathogen group) was much improved, with P values of less than 0.05. In model 2 of Salmonella spp., the concentration of Salmonella spp. had 96.6% of the data variation explained by ENT, HPyVs, and TN. In model 3 of norovirus GII, 76.2% of the data variation could be explained by the two variables, i.e., M. smithii and HPyVs. Hence, the multiple-linear-regression model had much better predictive capability (with higher r² values of >0.76), when alternative indicators (i.e., M. smithii and HPyVs) and TN were included, than traditional indicators alone.

DISCUSSION

Indicators.

By comparing the five sites from the reservoir catchment (i.e., sites 1 to 5), all of the sites except site 1 (inside the reservoir) had human-specific markers with concentrations ranging from 17 to 1,726 GC/100 ml and elevated E. coli and ENT concentrations, suggesting that the upstream catchments may be contaminated with human waste. A one-way ANOVA test showed that the mean concentrations of indicators from site 2 were significantly different from those at the other 4 sites (P < 0.05), indicating possible sewage contamination, perhaps from a leaking sewer. This might need further confirmation in the future.

In general, qPCR measurements of E. coli and ENT were approximately 1 log unit higher than corresponding culture-based results, even though positive correlations were seen between the results of the two methods (the r² value for E. coli was 0.388, and the r² value for ENT was 0.452). This suggests that the two techniques correlate well with each other in detecting E. coli and ENT in tropical surface waters, similar to previous studies (12, 23). However, the qPCR results correlated better with pathogens.

In addition to the traditional fecal indicator bacteria, the alternative indicators, F+ and somatic coliphages, were also detected in more than 90% of the water samples. This can be explained by other possible sources of coliphages, such as warm-blooded animals (24), since less than 50% of the samples were positive for human-specific microbial indicators. Moreover, somatic coliphages are able to replicate outside the human and animal gut when host and nutrient conditions are favorable (25). The results for the human-specific microbial indicators showed that HPyVs were detected most frequently, followed by B. thetaiotaomicron and M. smithii. The presence of these three indicators strongly suggests the possibility of human fecal contamination in the sampled waters. Since Singapore is completely sewered, the likely sources could be leaking sewer lines or illegal discharges from temporary structures (such as mobile toilets at construction sites). A previous study in Florida also showed that HPyVs generally had higher detection frequencies (5). This is not surprising, since HPyVs have widespread geographic distribution around the world and around 90% of healthy adults carry HPyVs (5). The occurrence of M. smithii in surface waters was low compared to that of other microbial indicators, consistent with other studies (26). This suggests that HPyVs could perform better as a fecal indicator, whereas the use of M. smithii alone may not be sensitive enough to detect human fecal contamination in environmental waters.

Correlations between different indicators.

The data from this study showed positive but moderate correlations between all of the traditional and alternative human-specific microbial indicators. Previous studies showed that positive correlations existed between bacterial indicators, whereas HPyVs showed negative correlations with traditional indicators (i.e., E. coli and ENT) and HPyVs were slightly more persistent in sewage (5). The differences from this study could be due to different death or removal kinetics for HPyVs (27). In sewage, high levels of nutrients and turbidity protect indicators of smaller size (i.e., HPyVs) more effectively, resulting in slower decay than traditional bacterial indicators, and thus leads to negative correlations (5). In our study of surface waters, the levels of nutrients and turbidity are much lower than their sewage concentrations, and hence, HPyVs may not have an advantage in survival relative to bacterial indicators. Thus, we observed positive correlations between all microbial-indicator studies, including HPyVs.

As expected, turbidity correlated positively with all of the microbial indicators. Comparing TN and TP, TN had more positive correlations with microbial indicators, suggesting that TN could be the nutrient limiting factor for microbes here, consistent with a previous study (28). DO and pH correlated negatively with almost all of the microbial indicators (except for coliphages), consistent with the anaerobic and relatively constant pH environment in the human gut. However, because somatic coliphages are known to reproduce outside the human gut under some aquatic environmental conditions where there are sufficient numbers of available host cells, this ability to infect and possible proliferate may make them less vulnerable to oxygen and pH changes in surface waters (25).

After heavy rainfall, the concentrations of all microbes decreased compared to dry-weather measurements, except for M. smithii, F+ coliphages, and norovirus GII. Rainfall can affect loading dynamics, either by dilution or indirectly by flushing microbes trapped in soils and sediments from previous contamination events (23). From our correlation analysis, only E. coli and ENT showed a moderately strong negative relationship with the duration of the dry period. This is consistent with studies by Wilkes et al. (29) but contrasts with others, which show that rainfall correlates well either with ENT but not E. coli (23) or with only E. coli (30). Singapore has a typical tropical climate, which favors the regrowth of E. coli and ENT in natural waters (31). Hence, it is difficult to conclude whether increases in E. coli are due to recent fecal contamination events or regrowth in the environment. However, if the human-specific markers (such as M. smithii and HPyVs) are detected as well, this increases the possibility of fecal contamination, because human-specific markers are not known to regrow in natural environments.

Correlations between pathogens, microbial indicators, and environmental parameters.

Correlations between indicators and pathogens depend on several factors, including transport and survival in the natural environment, sample size, positive percent detection, concentrations of pathogens in samples, land use, anthropogenic activities, and geographical location (32). For example, in temperate waters, a study by Ogorzaly et al. (33) showed that human adenovirus correlated with E. coli, ENT, and somatic coliphage, while another study by Horman et al. (34) did not find any significant correlation between indicators and pathogens. For subtropical waters, no significant correlations were observed between ENT, host-specific Bacteroidales, and pathogens (35). However, significant correlations were found between indicators and pathogens in this study. This could be due to the samples in the study originating from highly urbanized catchments with similar source characteristics and anthropogenic activities. Singapore has a high population density with few agricultural or animal sources, and hence, E. coli and ENT are likely to have originated from similar human sources. In addition, there were relatively high numbers of positive samples (from 20% to 92%) for pathogens, as well as microbial indicators. Prevailing eutrophic (high-nutrient) conditions in urban waters in Singapore (28) may also enhance the survival of certain microbial indicators and pathogens together, such as E. coli and Salmonella spp., leading to higher correlations between them (36).

In this study, there were better correlations between qPCR values of E. coli and ENT and enteric viruses (i.e., noroviruses and adenoviruses), while both qPCR- and culture-based methods for these indicators gave similar correlations with pathogenic bacteria (i.e., Salmonella spp. and P. aeruginosa). Hence, qPCR methods generally performed better as predictors of pathogens in this study. There are several possible reasons for these results. First, the conventional culture methods are sometimes unable to detect potentially viable target cells that are present, but qPCR is able to detect viable but nonculturable (VBNC) cells that the culture methods do not detect (12). In addition, the designed primer-probe set for ENT has high specificity for Enterococcus faecium and Enterococcus faecalis, which are abundant in animal gastrointestinal tracts, whereas the culture-based method is less specific and may grow nonenteric species that occur naturally in the environment, such as Enterococcus casseliflavus (37). Studies have also suggested that qPCR results are more capable of predicting gastrointestinal illness than culture-based results (38).

Among the pathogens, Salmonella spp. and P. aeruginosa showed better correlations with the microbial indicators than did the enteric viruses. Wu et al. (32) suggested that either more positive samples or higher population carriage rates would result in better correlations. Earlier studies showed that up to 10.3% of the U.S. population carried Salmonella spp. (39) and that approximately 4.3% of all nosocomial infections in the United States were caused by P. aeruginosa (40). For enteric viruses, the population carriage rates of rotavirus, astrovirus, norovirus, and adenovirus were 17%, 10%, 12%, and 8.2%, respectively (41, 42), and hence, the population carriage rates of different pathogens are similar. However, in the case of positive detections for our study, both Salmonella spp. and P. aeruginosa were detected in more than 90% of the water samples, which far outnumbered the positive detection of enteric viruses. Hence, it is possible that Salmonella spp. and P. aeruginosa had better correlations with microbial indicators than the enteric viruses due to the higher numbers of positive samples. Among the different indicators, the concentrations of E. coli and ENT from qPCR correlated better with the two bacterial pathogens than other indicators. This might be due to persistence and survival characteristics similar to those of Salmonella spp. in freshwater. For example, Salmonella spp., E. coli, and ENT were found to be able to survive in subtropical or tropical freshwater for more than 20 days (43, 44).

Norovirus GII was the most prevalent and had the best correlations with most of the microbial indicators, including E. coli (qPCR), ENT (qPCR), and the human markers B. thetaiotaomicron and M. smithii, while in contrast, F+ and somatic coliphages did not correlate well with enteric viruses (P > 0.05). This is opposite to one study in which F+ coliphages showed excellent correlation with adenoviruses (r = 0.99) (1) and another with positive correlation between norovirus GII and somatic coliphages (45). The poor correlation could be explained by nonhuman sources, such as birds, which cannot be discriminated by standard plaque assay methods, unlike the human-specific markers used in this study, which were also detected using qPCR. A better and perhaps more accurate method to predict enteric viruses would be to perform genotyping of male-specific RNA (FRNA) coliphages by RT-qPCR. One study has demonstrated significant positive correlations between human-specific coliphages (FRNA genogroup II) and human adenoviruses in a river from an urbanized watershed (33).

Among the pathogenic viruses, rotavirus and astrovirus are the only viruses that did not correlate well with any of the microbial indicators. A similar finding was also reported by Ferguson et al. (46). One possible reason is the different survivability characteristics between the viruses and microbial indicators. Lytle and Sagripanti (47) showed that the type of nucleic acid within the virus particle influenced the absorption of UV radiation and virus inactivation. Pyrimidine dimers, especially thymine dimers, are the most common lethal products from UV irradiation, and because thymine is present in DNA but not in RNA, this potentially makes RNA-containing viruses less sensitive to UV damage than DNA-containing viruses (47). In addition, it has been suggested that rotavirus has greater resistance to proteolytic degradation due to its capsid containing 3 layers of proteins (48). Thus, rotavirus may tend to have better resistance to UV inactivation and proteolytic degradation, and this may help to explain the lack of correlation between rotavirus and other microbial indicators. In the case of astrovirus, the lack of correlation with microbial indicators could be due to the fact that only 24% of the samples were positive for astrovirus.

Multiple-linear-regression analysis.

In the predictive models for Salmonella spp., ENT, together with HPyVs and TN, showed better predictive ability (r² = 0.966) than ENT alone (r² = 0.426). To date, there have been no studies on the relationship between Salmonella spp. and HPyVs. One study, however, did show that human-specific Bacteroides had significantly better predictive capability (r² > 0.56; P < 0.001) for the presence of Salmonella spp. than the conventional indicators (i.e., fecal coliforms; r² = 0.28) using logistic regression analysis (49). However, in our study, Bacteroides was excluded from the predictive model during the stepwise regression analysis, and other variables were automatically selected, with better contributions to the r² value. Relationships between pathogens and indicators have been suggested to be site specific (27), and our study was conducted in a tropical urban environment, whereas that of Savichtcheva et al. (49) was carried out in the northern temperate zone.

The predictive model of norovirus GII was also improved by adding M. smithii and HPyVs to the model, with the r² value increasing to 0.762. This can be explained by the fact that both M. smithii and HPyVs are highly specific to human wastes, with host specificities of 100% and 96%, respectively (26). Since GII noroviruses are human enteric viruses (19), it is not surprising that M. smithii and HPyVs have better predictive capability than the other indicators. In addition, higher population carriage rates would give higher correlations (32). HPyVs are distributed widely throughout the world, and their population carriage rate is more than 90% (5). M. smithii is also found in about one-third of the human population (13). In addition, because both HPyVs and noroviruses belong to nonenveloped virus taxonomic groups, they may have similar characteristics and survival patterns due to their similar sizes and structures. In fact, a recent study showed that HPyVs were able to mimic the fate and persistence of enteric viruses, where decay rates of adenoviruses and HPyVs were comparable but different from those of culturable bacterial indicators in sewage (5).

From the results of this study, both traditional indicators, E. coli and ENT (measured via qPCR), were confirmed to be good choices for predicting pathogens in tropical urban surface waters. However, if the human-specific indicators, B. thetaiotaomicron, M. smithii, and HPyVs, and environmental parameters (such as DO, nutrients, and turbidity) are also included, the predictive capability for detecting and quantifying human fecal contamination would be much improved. These predictive models could be applied to exposure assessment in quantitative microbial risk assessment (QMRA) to assess the human health risk in recreational waters. However, further study on the survival of both microbial indicators and pathogens in different geographical environments would help to validate these results and provide a more representative approach to exposure assessment modeling, which, coupled with dose-response models, could make QMRA a tool for recreational-water management.